Talks & recordings.

Hi, thanks for having me. It's very humbling to be among such a, such aesthetes and to be preceded by such aesthetes in particular. I'm still sort of reeling from that bitter taste and the sort of formaldehyde jar. But could we have the, All right, excellent. So I'm going to talk to you about some work on synesthesia and food, and using computers to compute synesthesia. My dissertation work is on taste, on modeling people's points of view and their cultural tastes by looking at things like their blog, their social network profile, and my hypothesis is a psychoanalytic one.

The idea is that if you like A and you like B, there's some fundamental correlation between A and B that you don't know, it's in our unconscious. And the same thing with food. A fun hypothesis I like to provoke people with is that anytime you're going through the day, you're looking at artwork, you're talking with people, anytime you're making any choices, that all impinges upon food, and it always suggests an underlying desire or... sort of an underlying longing or need for food.

And so the idea behind the synesthetic cookbook is that any idea, any emotion, color, sound, can suggest food. And so anything actually can suggest anything. So, I'm going to tell you about two particular things. The first is, what's for dinner? I want a machine to help me figure out what I should make for dinner, in terms of recommending recipes. And the question is, can a machine imagine how a recipe is going to taste? And the answer is, yes it can, if you tell it certain things.

A recipe has ingredients, procedures, context attached to it. And if you can tell it a little bit about each ingredient, each procedure, what the kitchen chemistry is, what the flavors and evocations are, then it can imagine how something will taste. And the power of that is that, not that it can replace aesthetes, it can't, but what it can do is, for 98% of the population, that sort of revere recipes is these monolithic units that cannot be changed at all.

What it does is, it gives them an opportunity to you know, behave, emulate the behavior of aesthetes, in that they can enter some keywords that they feel like, "I want something yummy, primal," and it will pop up some recipe suggestions because the computer is doing the imagining beforehand. And... So the second question is, can a machine behave as an aesthete and actually create recipes? And this requires some sort of data mining technology that's applied to a corpus of recipes, 160,000 recipes.

What's nice is that a recipe is basically a meme. Someone said, "These things taste good together under this context." And what I did was had the machine deconstruct a recipe and says, so that instead of regarding a recipe as something monolithic, it says, "Okay, well, these things together make sense. Okay, well, these ingredients together make sense. Well, what's the underlying fabric of things that make sense? How should recipes' ingredients connect to each other and influences connect to each other?" So I'm going to tell you first about how we've taught machines to imagine the taste of food.

And so there's a website called Open Mind Thought for Food, which in my lab, we've done a whole vein of work on common sense computing, as in having people type into the web things that everyone knows but no one's going to say because it's so obvious. And the machines don't know what a sort of a lemon tastes like. And so this interface is to acquire these sentences about food, like lemons are yellow, lemons have zest, lemons taste sour.

And people can type in these facts, like when eaten, ripe avocado usually has the palate texture, let's say, mushy. And you take thousands of these, tens of thousands, hundreds of thousands of these facts, and you're really getting out of these are the connections. You're getting an annotation that says, "Ripe avocado has these flavors." So this is what I call gustatory common sense, and it was acquired through this website. And we have about 54,000 descriptions of ingredients, cuisines, cooking procedures, just annotations which come to mind.

This is not, the people who are annotating this aren't, you know, aesthetes and critics. What they are are everyday people, but there is something emergent about their common sense. They share sensibility. And so, also to make aesthetic associations that are a bit more advanced, I basically had a computer algorithm read two volumes which I are very due to my heart the Oxford Companion to Food and the Food Lover's Companion and these have dictionary entries about Everything sort of under the Sun and what's nice is that you can mine?

Associations out of those so that anytime you think of let's say Asiago you think of something else, you know You think of feta you think of Greek and out of these associations we mined out 980,000 aesthemes which is an association strong association between two things and also 100,000 context because some things, your interpretation changes as the context changes, like Tim's experiment with the jar and the labeling between body odor and what was the other one, prosciutto?

Yeah, yeah. You got me on that one. So here are some of the synesthetic keywords that the system would be able to handle. And they're sort of very arbitrary. Like you can map music and sounds and smells and everything and artwork even into taste space. And these are statistically learned associations. And the idea is that one keyword is not going to give you the recommendation of what's for dinner, but it's the intersection of all of those and the latent connections that we don't consciously think about all the time that's really interesting.

Okay. So here actually is the interface, which is created with the help of a collaborator, Matthew Hockenberry. And it's basically a plate, and you type keywords into the top, like chicken, onion. You get these recipe results which fit that criteria. And after looking at some of the things, you might type some more, like spicy. So now you have spicy recipes coming up. And you see, oh, buffalo wings, I don't like wings. So type no wings. And as you see here, there's little critters here.

They're actually your family members. And they can program their own taste buds. And as you browse, you're going to get their feedback. So, here's that detail. Sally, so she can actually do this on IM. This is a nice interface. So I want rich, moist, spiced, I must have... "No nuts, not healthy, niacin." So it takes a few hundred thousand of these keywords. The idea is to accept everything, every word that can be spoken. And so the mom's browsing for something to cook, you know, "Niacin, niacin, I want niacin." And so these are picky critics.

But that's a point of view. That's really what it is. It's a set of contexts held together loosely and reacting against everything. So the second part is teaching machines to compose foods. which is very provocative for this crowd. So, recipe deconstruction. I happened upon sort of a corpus or a collection of recipes, 160,000 of them, and being that my background is in computational linguistics and artificial intelligence, I parsed these recipes. It's actually much easier to parse than just plain English because everything is laid out for you.

it's easy to normalize. And what I do is, rather than regarding each recipe as a thing to be revered, I say, "Okay, well, every time the machine sees a recipe, it remembers that these ingredients go well together, these procedures go well together in this order." And after looking at all 160,000 recipes, it's created the deconstruction. It takes... 24,000 unique ingredients mined from this corpus, and it groups it into 5,300 families of ingredients. Some are very easy to expect, like minced garlic goes with garlic cloves, but some are sort of very far off and far-fetched.

And also, the same is done with preparation techniques. That's collapsed into families. And it also associates ingredients and procedures with the cuisine that it could be indigenous to or make sense with respect to. So, statistically, you get a very interesting view of food. And what's nice is that I feel like sort of an archaeologist, and I'm excavating from these recipes all the good ideas had through, you know, recent history. And so these are the top 20 ingredients in this corpus.

You can see that it has a bias toward Americana and also toward baking. Unfortunately, there's not been able to parse foreign language recipes quite so readily. So, you have this frequency chart, and it's interesting because just like language, the distribution of ingredients is exponential. As in, like, the most frequent, like the word "the" is used so many more times than, like, you know, the third or fourth most frequent. So let's say you look at Italian cuisine, and these are the relative significances of these ingredients in Italian cuisine, and you can see that it sort of, you know, not really, but it sort of maps the contour, the general contour, where salt and sugar are the most prevalent.

But then you look at another profile, let's say Chinese cuisine, and you can see it's completely different. And what's interesting is that if you could take every cuisine and you could reduce it to this list of frequency bars and histograms, that would give you an incredible amount of information already about what the taste profile of that cuisine is. So I said, you know, the computer looks at every recipe. Every time it sees two ingredients together, it says, okay, I'll remember that.

And every time I see those two ingredients repeatedly, I'll reinforce that association. So what's nice here is that this is a self-organizing map of correlations between things. So I paused this for a second to show you that... So every link means that these things go well together. They complement each other. Like, you know, canola oil, minced ginger, baking soda. But what's nice is in a self-organizing map in a network is that when you... You say these two things go together well, these two things go together well, and you put them into a map, it organizes and it creates cliques.

And you can actually see that this is sort of like the baking clique, or, you know, French bordering up over here. And if you keep going on this side, we can see another clique. Over here we have some spices, right? And we have more dry spices over here. And I think this goes a little bit further. Yeah, so you have on this side bordering Indian or Mediterranean cooking on this side. So that's an example of how, if you mine things like in this way, you get a sense of emergence, which is something that you didn't know could be had, just that there are fundamental relationships between ingredients.

Okay. So here's another one. And so I like to provoke you with a thought that authenticity is really an aesthetic closure. It says these things go together well. You leave out these things which would vulgarize this thing. And these things are, they make sense with respect to each other. They form a clique. Okay. And so what this visualization is showing is that, so you have this middle node called Who Am I? And it's just collecting things.

And I've typed three ingredients to collect around Who Am I? So it's like a guessing game. I'm thinking of a cuisine. It has cumin, turmeric, bay leaves. And all the yellow notes here are the other things it predicted that would be in the same recipe. It predicted chili powder, you know, oregano, carrots, paprika. So it's got the general sort of region, right? And then using a similar approach by comparing cuisines, like, oh, this ingredient is in this cuisine, but it's not in this cuisine.

we extracted some essences so here are actual top results for essence of Thai, essence of enchilada and you can actually get essence of any word just by looking at all the ingredients which collect around that word So here actually is a recipe composer. It's called Gulp Fiction. And the idea is that, so I typed the word blue, and this is synesthetic because it's got all of it. And it actually, it's writing the recipe as I type it.

Blue, it's got blue cheese, mayonnaise, sour cream, you know, white wine vinegar, Worcestershire sauce. So it seems like it's looking for acidity, but it's not instructed that explicitly. And it also tries to invent a little procedure for cooking it, which is abstract at best. So now you type another word, blue and sudden, and now what it's done is it's intersected blue and sudden. So what happened here? Well, you know, you got lime, pineapple, maybe some more pleasant things, because sudden could be pleasant.

You got some broccoli, um... And the procedures change. And then you type the final word here, "salsa." And so you have -- all right, so now you have something which is kind of salsa-y. The little cheat here is that the way the recipes are named, generally the last word is the most salient, so you have to give it the most weight. So you have lime juice. You have a very interesting salsa because it has blue cheese.

But, you know, in the context of this crowd, this is art and not garbage, right? So here's another one. We have, and this is, we're going completely off the deep end here. Mediterranean, which, you know, this is a typical profile. I got balsamic vinegar, olive oil, aubergine, etc. And also, the aubergine and the eggplants are, you know, obviously, just like the erycoverian green beans, have completely different contexts, just because they contain sort of a lot of contextual information, even though some people would argue that the taste profiles are pretty similar.

So we have Mediterranean, and then we're going to disrupt that, and you're going to throw in caveman. So, what it's done really is that we're making beef short ribs, because apparently that's very primal and courgettes, but it's Mediterranean because it's got feta and it's got capers and balsamic vinaigrette. So, by the way, if any of you have restaurants, would like this cooked, be happy to entertain this. I had the pleasure of doing a couple of these food performances on the invitation of some folks.

And I think one group, there were artists that were invited to this dinner, and they gave words such as "misunderstood," "gravel," whatnot. And the strange thing was the intersection of all those was like a very strange black bean porridge. It was very dark. So here are some recipes around paella. What I wanted to tell you is that, you know, we're all aesthetes here. We're in the 2%. And in 98%, what would it mean to someone if you gave them a recipe for paella and then you allow them to modify the paella by adding other influences.

Here I've added, you know, this is Thai paella, so I've thrown in fish sauce. And by the way, the quantities are all inferred also because there's a very primitive sense of kitchen chemistry which happens in this recipe. maker and um and coconut milk um so that's sort of plausible then you have shepherd's paella which means you throw in you know mutton and uh apparently uh and uh it could be could it could work could work If it was served at one of your restaurants, it would definitely work.

Then you have something more implausible like a paella compote, which, you know, dried apricots can work. But, you know, what I think of this is it's sort of a training ground for asthites, aspiring asthites, because they can experiment with different taste profiles, mixing them together. And lastly, paella Suzette, which... I'm not sure if this is dessert mussels with blueberries and cranberries. Yeah. And what I wanted to leave you on was my discovery of the fact that, based on all these ingredients and all the combinations that these ingredients typically fall in, that only represents about 2% of the possible combinations that ingredients could be in.

So that's sort of, you know, 98% to go. That suggests a very promising future for the field of food science. So thanks very much.

Machine transcript (WhisperKit, large-v3 turbo, on this machine, 12 September 2026), corrected only where a name or term was misheard; the player’s captions carry the timings.

Please tell us about your position at eBay and your area of expertise.

Yes, well until very recently I was a principal scientist at eBay Research Labs and I was conducting research using data science to study brands and tastes

Okay, what is your definition of taste?

Yeah, sure. So taste is both something that you experience in your mouth, on your palate, but also it's a larger concept. It's about your system of style and aesthetics. And I think that taste in our day and age, in the age of the internet, is a core part of our identities. It's the part of our identities that we can explore and control the most through the consumption of media and experimentation.

How predictable are we? Is taste something that can change or is it a constant for each individual?

Well, for me it's changed a lot. And I think the more time and effort that you invest in understanding yourself and attempting just experimenting with clothing yourself with with different tastes trying things on and looking around the more you can grow and it's a process of growth and self-discovery

What tools and models do you use to collect data about taste and to visualize it?

At eBay we have a big data practice where we store and analyze data about customers and transactions so we can improve the user experience so we can make content more engaging for users and more relevant and and if the user wins then the business wins too at MIT you know we had our own sort of methods for collecting taste data and I I use a family of approaches pioneered by Pierre Bourdieu, the French sociologist, of factor analysis as primary tools in the analysis of taste data.

And it's a bit science but also a bit art because there is some interpretation in there. And I also apply some theory from sociology.

Is there a danger that when marketers use this data that we get stuck in a bubble if we are only offered products that we should like according to our taste?

Yes, so in my presentation I showed that there are early adopters of innovations and one of their core behaviors and I think is a virtue and a lesson for everyone is that the more time and effort you invest in understanding your own tastes the harder it is to be manipulated in this way. and really connoisseurship and growing your tastes is really about exploration, self-criticism, self-reflection. And with that, I'm not so much afraid that for active consumers, you may get stuck in a taste bubble.

I do think that a lot of passive or casual consumers may be a victim of taste bubble or just in general information bubbles, filter bubbles.

For a company using taste data, what is the most important things to think about when targeting their audience?

Well, always make sure that you are doing right by the customer and the customer experience should always be the motivating force and the driving factor. for the use of this and it certainly was for me at eBay because our goal was to create more engaging, compelling, relevant content by understanding our customers' tastes. So my advice is always be motivated by improving the customer experience.

Machine transcript (WhisperKit, large-v3 turbo, on this machine, 12 September 2026), corrected only where a name or term was misheard; the player’s captions carry the timings.

Alright, I asked each of you to pick iceberg and arugula, and you have those in your hands. Now I'd like you to pick your favorite of the two and raise it up. And I will tell you something about yourself. All arugula, oh, one iceberg, okay. So if you picked arugula, it means statistically you're more likely to be politically liberal.

To prefer rucola can mean that you are more liberal than the one who prefers iceberg salad. That's what taste researcher Hugo Liu means. By analyzing the choices we make, he can help companies understand what consumers want, why we want it, and perhaps most importantly, what we will want in the future.

So my work, my PhD work was on computer models of taste, and it's based in the work of a French sociologist named Pierre Bourdieu. He made these wonderful taste maps using French lifestyle surveys from the 1960s. And he was really the first social data scientist of the modern age. I'd like for this for you to pick up either a jar of crunchy peanut butter or a jar of smooth peanut butter. Okay, so we have crunchy, smooth, crunchy, smooth.

Okay, so if you chose crunchy, statistically you're more likely to be personally liberal. So I want to explain to you why I did this demonstration with you guys. The example of lettuce and the example of peanut butter, they show the power of using statistics to predict your tastes so that we can customize experiences to you and help you discover things that you love.

I have a feeling that there's great money in this kind of business, helping brands knowing how to get to their customers.

How much do you get paid?

You know, I think it's important for technologies to be adopted. And for that to happen, people are going to make money. Whether it's me or someone else, I don't know. Thank you. Thank you.

Machine transcript (WhisperKit, large-v3 turbo, on this machine, 12 September 2026), corrected only where a name or term was misheard; the player’s captions carry the timings. The programme's voice-over is Swedish on the tape and is rendered in English by the transcriber; Hugo's own words are in English.

Thank you. All right. So when we pass the keywords around -- when we went around for talk slides, I thought of a keyword and it was just a metasyntactic variable, but you guys wanted it, so here Minimum viable lifestyle. All right, so one of my favorite quotes from Coco Chanel is, when accessorizing, always take off the last thing you put on. And I think that's very true because people do too much. They do more than they need to in their life.

They think that everything is helping them to accomplish. their goals but that's not true so the question is you know what are the accessories you should take off of your life so I'm a big believer in the Nietzschean becoming in the velocity of iteration of self and I think that this just like you know your access to a gym is proportional to the inverse of the square of the number of features or baggage you carry in your life So we want to nudge you toward parsimony of features in your life so that you can iterate and become a better person.

Alright so I'm just gonna offer you ten tidbits because this is a five-minute talk. A crowd that wears black. So if you had a crowd that wears all black and standing out outside a nightclub and a crowd of people wearing sort of all colored clothes, I would challenge you that the people who were wearing all black, they're more individualistic, they're more interesting, and actually much more diverse. Oh, there you go, okay, so. Thanks for recognizing that, Lily.

So my concept for you, my teaser is, everyone has the same size feature space polygon count in life. And I would say that monochromats like myself have an advantage because we economize over color space. We don't blow all of our features on our wardrobe, okay? So... Also, if you wear black, nobody is going to notice if you wear it over and over again. That's very helpful in life. Denim. If you have denim, you don't need to be washing it all that much, or if ever at all.

A.P.C., a French fashion brand, they're very clever. They have a line of denim called Butler jeans. So the idea is you can wear something out, raw denim, for six months and then turn it in for a huge discount. And someone else will pay four or five figures for your worn-in denim. And the idea is that they say in the Middle Ages, people used to not wear their own chain mail. They used to get butlers for them to wear in their clothes for them.

And it's really nice because once you realize you don't actually need to be washing your denim all that much, you don't need to be owning all that much denim. And, in fact, I don't. Okay. So... I traveled, I travel a lot. I like to travel to, you know, Southeast Asia for two weeks at a time on a very, very small, you know, 10 liter backpack. And people are like, well, is that sanitary? You know, can you get away with it?

And you have to realize that like packing intelligently is very important. And underwear and undershirts are the limiting reagent of your travel. You don't have to pack multiple pairs of pants, multiple shoes. Your toiletries bag can be very small, but you need a lot of underwear. So I'm not asking you to economize on underwear. Just remember that. The shampoo and conditioner industrial complex. I was not drinking when I read this slide. I mean, so the story is I used to use shampoo, and no longer for over a year I haven't used shampoo.

And think about it. You wash your hair, your hair gets dry and brittle, then you use conditioner because your hair is dry and brittle and your hair gets nice and moisturized for five seconds, but then over time your hair gets oilier, you wash your hair more, you rinse, repeat. because you need it. And so they've created this fictive product, inserted this need in your life that didn't need to exist at all. Because most people say, oh, when I stop shampooing, my hair gets really oily.

But that's just a recoil effect. So just get over it. Especially if you're a guy, go for it for a few weeks, and you'll see that you don't need shampoo and conditioner. You just need some hot water to remove some of the excess oil. and there's a lot of reasons to avoid shampoo sodium lauryl sulfate being one of them it's a surfactant it's very bad for you if there's sodium lauryl sulfate in your toothpaste also it will give you canker sores stop using it sports you don't need to watch it because I used to watch sports and after I stopped I got 20% more time back All right, so the hegemony of recipes.

Recipes are really, really complicated, and not for, I think, for pretty arbitrary reasons. I have a lot of chef friends, and they confess to me that they have written recipe books, and they have recipes that they've never cooked. So why are you following that? So I think that the way to do it is to cook for yourself. It's much more delicious. Cook single vegetables. Take some celery, try to steam it, saute it, braise it, do whatever you want.

Understand the vegetables. Build a personal relationship with vegetables like you do with Jesus. Yeah. All right. Prophylactic panaceas. I was on an alliteration kick. So I'm a big believer in panaceas because they are very economical. One of my favorites is saline. So if you feel like you're getting sick, you can irrigate saline. your sinuses, your ear canals, your throat with it. I also use saline to style my hair because as you know, the Aussies figured out that they call it surf spray, that surfers have really nice hair.

Well, saline is a great texturizer. Take a really good multivitamin. One way to tell how good your multivitamin is beyond the BS, the marketing BS, is look at where, what the source of vitamin E is. There's actually eight or nine forms of vitamin E, and the cheapest one is alpha-tocopherol. That is synthetic, it's from petrochemicals, and you don't want that one. So you want the one that has, the best distribution of vitamin E's. There's a lot of other panaceas, like adaptogens, ginger, turmeric, holy basil.

Talk to me about it. Weak ties. There is a really insidious paper in the Journal of Sociology by Mark Granovetter in 1973 that everyone cites that's... the strength of weak ties, okay? So I think people got the misimpression that they should be networking. I personally hate networking, and I think that it's much better just to make real friends, like friends that you really want to hang out with, and to stop networking, because the few strong ties you have, if you weight all of your connections by the strength of your ties, like strong ties will get you a lot further, friends.

The metaphysics of presence. Something that my friend, my good friend Linda Stone cued me into was our common therapist who is a an existential investigator. And she helped to cue me into something that also I read from a book called Zen and the Art of Motorcycle Maintenance, which is this idea, this quality of being present. Not in the millisecond before, and not in the millisecond after, but being present because that is like a news feed, a better Facebook feed than Facebook.

You should be in on this. and this is like really, really good stuff to read. Campari. So I used to, I feel like, be a functional alcoholic and then I discovered that white wine masks kind of alcoholism pretty well. But I found that another drink was more economical than white wine even, which is Campari. So you start with a Negroni, which is Campari, gin, and vermouth. You can step down later in the night to an Americano.

Then you can step down to a Campari soda. Something lighter would be an Aperol soda, but all very much like Campari. I like these alcohols that you can build a whole spectrum of drinks out of. Anyways, that's an arbitrary deck for you.

Machine transcript (WhisperKit, large-v3 turbo, on this machine, 12 September 2026), corrected only where a name or term was misheard; the player’s captions carry the timings.

Hi MoMA. My name is Hugo Liu. I'm the chief scientist of the online art marketplace, Artsy. Paola, thank you so much for a very provocative prompt. Is art the real Turing test? For me, no. it's not. In a sense, I think that art is maybe the worst Turing test. In the Turing test, the examiner is adopting a skeptical stance, attempting to throw curveballs in a conversation with an examined black box. In an attempt to expose the examined black box is untrue, whereas in contemporary art, on the other hand, Contemporary art demands that the examiner adopt a stance of surrender, being completely receptive to the possibility that what's examined is actually meaningful and true.

So consider Marcel Duchamp's Fountain and his other ready-made objects, for example. In the contemporary art game, the assumption is that what is examined is true. If anything, the examiner can only discover their own insufficiency in the process. However, I think we can strengthen the prompt with a friendly amendment. Is art interpretation the real Turing test? And to this I say possibly so. Contemporary art is always in dialogue with culture. To return to the example of Duchamp's fountain, the fountain is art in the sense that it was a potent intervention into the culture of that time, identifying a tension about what art is and exploiting it.

To explain why Duchamp's fountain is art means explaining how that work sits in dialogue with culture. This would require, I believe, a lot of context about where civilization, society, the human condition are up to the point, the current moment. It requires an understanding of the zeitgeist, the current moment. And in computer science terms, you can think of the current moment as a function like a hash or a succinct summary of everything that's happening right now. This, I believe, would be incredibly hard for an AI to grok the zeitgeist, and therefore it would make for a very good Turing test.

Thank you.

Machine transcript (WhisperKit, large-v3 turbo, on this machine, 12 September 2026), corrected only where a name or term was misheard; the player’s captions carry the timings.

Hello, it's Sunday, December 11th, 2016. My name is Hugo Liu, and I have a PhD in taste from the MIT Media Lab. I'm the founder and CEO of an art technology startup called ArtAdvisor. We're using cultural analytics to help collectors discover contemporary art. So I think maybe a month ago, my dear friend Jeremy Fisher, who had taken part in a previous Reality Tech, he strongly encouraged me to apply to this. And Jeremy is a very dear friend and he's a bit of a Sherpa for me because he has really good taste.

And so I pay really close attention to when he nudges me to apply for something. So after reading a little bit more about Reality Tech and looking at last year's itinerary, I got really excited to apply. I think I'm an ideal candidate for this leadership trip because my passion is the intersection of technology and cultural scholarship. I think Israel is such an old and influential culture. It has such a powerful energetic footprint, almost like a mystical footprint, that I think really getting to experience it firsthand would help me to deepen my understanding of the nature of all human cultures.

and more practically, my startup is machine learning focused and there's so much data science being done out of Israel today that I think it would be tremendously helpful for me to understand if there are some opportunities for us to partner with firms or universities in Israel to build out our company for the next 10 years. All right, that's all for now. Thanks for listening, bye.

Machine transcript (WhisperKit, large-v3 turbo, on this machine, 12 September 2026), corrected only where a name or term was misheard; the player’s captions carry the timings.